Sphinx eager ".. only::" directive and other selective rendition extensions

Project home page: https://github.com/pfalcon/sphinx_selective_exclude

The implementation of ".. only::" directive in Sphinx documentation generation tool is known to violate principles of least user surprise and user expectations in general. Instead of excluding content early in the pipeline (pre-processor style), Sphinx defers exclusion until output phase, and what's the worst, various stages processing ignore "only" blocks and their exclusion status, so they may leak unexpected information into ToC, indexes, etc.

There's multiple issues submitted upstream on this matter:

• https://github.com/sphinx-doc/sphinx/issues/2150
• https://github.com/sphinx-doc/sphinx/issues/1717
• https://github.com/sphinx-doc/sphinx/issues/1488
• etc.

They are largely ignored by Sphinx maintainers.

This projects tries to rectify situation on users' side. It actually changes the way Sphinx processes "only" directive, but does this without forking the project, and instead is made as a standard Sphinx extension, which a user may add to their documentation config. Unlike normal extensions, extensions provided in this package monkey-patch Sphinx core to work in a way expected by users.

eager_only

The core extension provided by the package is called eager_only and is based on the idea by Andrea Cassioli (see bugreports above) to process "only" directive as soon as possible during parsing phase. This approach has some drawbacks, like producing warnings like "WARNING: document isn't included in any toctree" if "only" is used to shape up a toctree, or the fact that changing a documentation builder (html/latex/etc.) will almost certainly require complete rebuild of documentation. But these are relatively minor issues comparing to completely broken way "only" works in upstream Sphinx.

modindex_exclude

"only" directive allows for fine-grained conditional exclusion, but sometimes you may want to exclude entire module(s) at once. Even if you wrap an entire module description in "only" directive, like:

.. only: option1
    .. module:: my_module

    ...

You will still have an HTML page generated, albeit empty. It may also go into indexes, so will be discoverable by users, leading to less than ideal experience. modindex_exclude extension is design to resolve this issue, by making sure that any reference of a module is excluded from Python module index ("modindex"), as well as general cross-reference index ("genindex"). In the latter case, any symbol belong to a module will be excluded. Unlike eager_only extension which appear to have issued with "latexpdf" builder, modindex_exclude is useful for PDF, and allows to get cleaner index for PDF, just the same as for HTML.

search_auto_exclude

Even if you exclude soem documents from toctree:: using only:: directive, they will be indexed for full-text search, so user may find them and get confused. This plugin follows very simple idea that if you didn't include some documents in the toctree, then you didn't want them to be accessible (e.g. for a particular configuration), and so will make sure they aren't indexed either.

This extension depends on eager_only and won't work without it. Note that Sphinx will issue warnings, as usual, for any documents not included in a toctree. This is considered a feature, and gives you a chance to check that document exclusions are indeed right for a particular configuration you build (and not that you forgot to add something to a toctree).

Summary

Based on the above, sphinx_selective_exclude offers extension to let you:

• Make "only::" directive work in an expected, intuitive manner, using eager_only extension.
• However, if you apply only:: to toctree::, excluded documents will still be available via full-text search, so you need to use search_auto_exclude for that to work as expected.
• Similar to search, indexes may also require special treatment, hence there's the modindex_exclude extension.

Most likely, you will want to use all 3 extensions together - if you really want build subsets of docimentation covering sufficiently different configurations from a single doctree. However, if one of them is enough to cover your usecase, that's OK to (and why they were separated into 3 extensions, to follow KISS and "least surprise" principles and to not make people deal with things they aren't interested in). In this case, however remember there're other extensions, if you later hit a usecase when they're needed.

Usage

To use these extensions, add https://github.com/pfalcon/sphinx_selective_exclude as a git submodule to your project, in documentation folder (where Sphinx conf.py is located). Alternatively, commit sphinx_selective_exclude directory instead of making it a submodule (you will need to pick up any project updates manually then).

Add following lines to "extensions" settings in your conf.py (you likely already have some standard Sphinx extensions enabled):

extensions = [
    ...
    'sphinx_selective_exclude.eager_only',
    'sphinx_selective_exclude.search_auto_exclude',
    'sphinx_selective_exclude.modindex_exclude',
]

As discussed above, you may enable all extensions, or one by one.

Please note that to make sure these extensions work well and avoid producing output docs with artifacts, it is IMPERATIVE to remove cached doctree if you rebuild documentation with another builder (i.e. with different output format). Also, to stay on safe side, it's recommended to remove old doctree anyway before generating production-ready documentation for publishing. To do that, run something like:

rm -rf _build/doctrees/

A typical artificat when not following these simple rules is that content of some sections may be missing. If you face anything like that, just remember what's written above and remove cached doctrees.